Agents that never leave your tenant: what Ignite 2026 means for data sovereignty
T

he arms folded question
Every enterprise AI conversation eventually arrives at the same question, usually asked by someone from security with their arms folded: “So where does our data actually go?”
At Microsoft Ignite this week, Microsoft gave that question a much better answer. Foundry Agent Service is adding deployment to customer owned AKS clusters, with a public preview targeted for early December. In plain English: you use Microsoft’s agent tooling, but the agents run on infrastructure you operate, inside your own network.
For regulated businesses, that is not a footnote. It is the difference between “interesting pilot” and “approved for production”.
Why sovereignty matters more for agents than for chatbots
A chatbot answers questions. An agent reads your contracts, queries your CRM, drafts your tender responses and, increasingly, takes actions. The more useful an agent becomes, the more of your sensitive data it touches.
That is why the hosting question gets sharper, not softer, as AI matures. In financial services, FCA expectations on outsourcing, operational resilience and accountability all apply. Under GDPR, you need to know where personal data is processed and by whom. In construction, tender pricing and client contracts are sensitive enough that nobody wants them taking the scenic route around the internet.
Put simply: an agent you can’t locate is an agent you can’t govern.
Where your data lives, and where it never goes

Everything an agent needs, from your documents to the model doing the reasoning, sits inside your own Azure tenant. Nothing in this picture requires client data to leave it.
What Microsoft’s announcement does, and doesn’t, change
It does make customer hosted agents a mainstream, Microsoft backed pattern. If your board has been nervous about agents running “somewhere in the cloud”, this is a strong signal that the industry is moving their way.
It doesn’t govern anything for you. Running agents on your own cluster means your team operates the cluster, networking and capacity. You still need to decide which processes deserve an agent, who owns each one, what it may touch and how its work is reviewed. Hosting solves the “where”. It does not solve the “what”, the “who” or the “should we”.
It is also a preview. Preview features are for testing, not for running your Consumer Duty complaints process next Tuesday.
How Neologik does this today
Here is the slightly smug part: Neologik’s platform already works this way. It is a governed AI workflow layer that runs in your own infrastructure, connecting your AI, data and systems without your data leaving your boundary.
In practice that means:
Your data stays put. Agents work with SharePoint, Dynamics and Microsoft 365 where your data already lives, under the access controls you already manage.
Governance is built in, not bolted on. Every agent sits inside a defined workflow with a named owner and human review points, aligned to FCA, GDPR and ISO standards.
One agent at a time. We start with a single high value process, such as RFP responses, audit preparation or contract review, prove the value, then scale.
We stay. As a certified Microsoft partner, we help you adopt new capabilities like those announced at Ignite as they mature, rather than leaving you with a tool and a handshake.
The bottom line
Ignite 2026 confirmed what regulated businesses have been saying for a while: the best place for an enterprise agent is inside the enterprise. The question is no longer whether you can keep agents in your tenant. It is whether you want to build and run that yourself, or start with a platform that already does.
Talk to us about deploying AI inside your own Azure tenant. Book a 15 minute conversation and we’ll show you exactly where your data would live, and where it never goes.



Comments